August 2018
Intermediate to advanced
272 pages
7h 2m
English
MobileNets has some of the best accuracy, speed, and parameter ratios for any neural network design.
However, currently there is no good (fast) implementation of depthwise convolutions for running on a GPU; as a result, training will likely be slower than using a normal convolution operation. However, where this network really shines at the moment is in small CPU designs, where the increased efficiency is more visible.
Read now
Unlock full access